To The Moon

Hello. The weekend is over, and the most striking thing left from Friday's close was a pair of numbers pointing opposite ways: $Advanced Micro Devices(AMD)$ up 6.50 per cent, $Broadcom(AVGO)$ down 5.94 per cent. Both are AI chip names, and they spent the day going in different directions.

What the market was reading that day was not results. It was second-quarter position disclosures. Stanley Druckenmiller's 13F showed him loading up on Amazon and AMD while dumping a batch of chipmakers, and a Soros fund was adding AMD too, alongside $Micron Technology(MU)$, $Applied Materials(AMAT)$ and $Taiwan Semiconductor Manufacturing(TSM)$. AMD's 6.50 per cent was booked to all that.

But AMD had business of its own that day. It had just completed a US dollar bond sale of about US$4.75 billion, reported as the largest in its history, and coverage hung the day's gain directly on it; Bank of America raised its server CPU market estimate to more than US$210 billion over the same stretch and named AMD its top pick.

Broadcom's 5.94 per cent cannot be booked the same way. Coverage puts it down to two things: a serious security exploit in VMware, and anxiety about how the AI buildout is being financed — one outlet framed the latter as a "US$370 billion financing question". The fund that exited Broadcom, along with $NVIDIA(NVDA)$, was a different one, Third Point.

One day, two directions, both explained by a single 13F. But they only happened to fall on the same day; neither drove the other. Each side had its own business that day.

The disclosures themselves are the more interesting part. A 13F records what was held on 30 June, not what was held on Friday. And within the same batch of filings, a Soros fund added to Micron, Renaissance cut it hard, and PRIMECAP trimmed it — the same stock, the same quarter, several of the best-known institutions cancelling each other out.

So the trouble with following 13Fs is not that they are a quarter late. It is that they do not point anywhere.

Something else that day did point somewhere. $Micron Technology(MU)$ closed up 2.30 per cent, a fourth straight session higher, the four gains running 0.87, 4.92, 4.23 and 2.30 per cent. The catalyst had nothing to do with 13Fs: New Street upgraded it to Buy with a US$1,250 target, on one argument — that memory has stopped being a cyclical story.

The same day $SanDisk Corp.(SNDK)$ closed up 7.39 per cent at US$1,641.11, up 35.38 per cent over the week. JPMorgan upgraded it after its investor day and put a US$2,250 target on it. The chain came along: $Western Digital(WDC)$ rose 4.41 per cent, $SK hynix(SKHY)$ 0.40 per cent and $Roundhill Memory ETF(DRAM)$ 0.69 per cent.

The interesting detail is that SanDisk told investors at Thursday's event that margins would fall, and the stock rose 13.67 per cent on the day anyway. What was being bought was not the current quarter. It was the number out at 2030: an 80 per cent gross margin.

The two upgrades rest on the same reason, and it is not the results: memory has been priced as a cyclical for years, and an 80 per cent gross margin is a number a cyclical valuation has no room for.

When a business stops being valued as a cycle and starts being valued as a grower, the rally is the gap between the two numbers. That gap has not closed, which is why both sides can still find evidence.

The bear evidence is specific enough. The company itself said margins would fall. Michael Burry added again on Friday to shorts in $Micron Technology(MU)$, $NEBIUS(NBIS)$ and $Oracle(ORCL)$.

And there is one thing at SanDisk still to be delivered on: the company says roughly half of its financial-year 2027 capacity and about two-thirds of financial-year 2028 capacity is already contracted to customers. If that number holds, "cyclical" gets harder to pin on it — a business that can sell its capacity two years ahead is not haggling over the price on the day.

Optical interconnect was Friday's strongest branch. $Applied Optoelectronics(AAOI)$ surged 15.53 per cent, $Lumentum(LITE)$ rose 5.19 per cent and Coherent slipped 0.43 per cent. AAOI had no news of its own — the only explanation on offer is that the whole optical group was moving, and that move was set off by Lumentum's upbeat results last week.

The other side of that is on the table too: a well-known short seller argues that commoditised bottlenecks do not make good companies. And $Coherent(COHR)$ beat on both lines after Wednesday's close, then closed down 7.99 per cent on Thursday — it has just shown how badly this chain takes any hint of price cuts.

$NEBIUS(NBIS)$ closed up 8.88 per cent, still riding its second-quarter numbers — revenue up 454 per cent — with the rally stretching from the release all the way to Friday. There was no rating action that day, only a Bank of America tweak to its 2027 target; a few days earlier DA Davidson had cut its target by 30 per cent. Burry's short list has it too. CoreWeave closed down 0.97 per cent.

The mega-caps barely moved. $NVIDIA(NVDA)$ fell 0.06 per cent, $Microsoft(MSFT)$ 0.30 per cent, $Alphabet(GOOG)$ 0.12 per cent, while $Apple(AAPL)$ rose 0.22 per cent and $Tesla Motors(TSLA)$ 0.68 per cent; $Palantir Technologies Inc.(PLTR)$ closed down 2.78 per cent and $Intel(INTC)$ 1.97 per cent.

The indices stayed put as well. $S&P 500(.SPX)$ closed down 0.17 per cent at 7,785.76, a day after a record close, and was still up about 0.4 per cent over the week; $Invesco QQQ(QQQ)$ fell 0.14 per cent. Rate futures now put the odds of the Fed holding in September at about 67 per cent — which is the same as roughly 33 per cent on a rise, the figure that was in the market the day July CPI landed on Wednesday.

$SpaceX(SPCX)$ closed down 0.91 per cent, and the disclosures on it came one after another. Nvidia's own filing showed a SpaceX stake of about US$21 billion, its second-largest equity holding. Harvard disclosed US$2.2 billion, Norway's sovereign wealth fund US$1.2 billion and the mining magnate Gina Rinehart US$1.4 billion, while Tiger Global cut Big Tech and bought into it.

The most striking was $Alphabet(GOOG)$: an investment of about US$900 million is now worth US$94 billion.

On 4 August SpaceX said on its earnings call that it was going all in on Nvidia for chips. Now the other end is visible too: $NVIDIA(NVDA)$ has US$21 billion riding on that customer. A supplier holding equity in its customer, and the customer committing to buy all its chips from that supplier — this is the circular financing the market has been asking about all month, and it now has a figure attached.

So 13Fs are not useless. They are being used for the wrong job. Read them for trade ideas and the best-known institutions cancel each other out on the same stock; read them for structure and one filing exposes the rope tying Nvidia to SpaceX.

With the first unlock of the largest IPO on record just behind it, that much money is aboard and the price still hardly moved — it was already in there, so none of it counts as new buying.

The week ahead is crowded. The minutes of the July policy meeting come at 2am Singapore time on Thursday: rates did not move a single basis point at that meeting, but three members argued for a rise, and how the minutes weigh inflation against growth decides how high growth stocks can be priced. Three retailers report pre-market — Home Depot on Tuesday, Target on Wednesday, Walmart on Thursday — where the reads are same-store sales and second-half guidance.

Hong Kong and the China names are bunched together too: $XIAOMI-W(01810)$ and $BIDU-SW(09888)$ report on Tuesday, $KUAISHOU-W(01024)$ after the Hong Kong close on Wednesday, $BABA-W(09988)$ pre-market in the US on Thursday. The tape has moved first — in today's Hong Kong session Xiaomi rose 1.80 per cent and Kuaishou 0.25 per cent, while on Friday in the US Alibaba rose 1.35 per cent and Baidu fell 0.96 per cent, the news on Alibaba that day being that Apple is training a China-specific AI model with its help.

$TENCENT(00700)$'s last set of results is the reminder: revenue up 11 per cent and adjusted net profit up 9 per cent, and the shares still fell more than 3 per cent on the day because capital expenditure climbed to RMB52.78 billion and free cash flow turned negative.

That 30 June snapshot can only tell you where the money stood on 30 June. What actually changed a price on Friday was one sentence in two upgrades: this business should no longer be priced on the cycle.

The above is personal analysis, not investment advice.

💬 【Talking Point】

Rate futures now put a September hold at about 67 per cent, and the July minutes land on Thursday — from a meeting where three members argued for a rise. Do you position for the minutes ahead of time, or wait until they are out?

💰 【Bounty】

Drop your view in the comments and there are coins in it for you! 🎁

🔔 Better shared than saved — tag a friend and split the coins!